A project on a convolutional neural network image classifier that classifies images of leaves based on their type and health.
This project involves the development of a convolutional neural netrwork model that detects the presence of diseases on leaves. The training data consisting of around 20,000 images is preprocessed and augmented. Tensorflow and Keras are used to load the data and build a convolutional neural network.
The neural network consists of three convolutional layers, each having a 2D convolutional layer, a maxpooling layer and a dropout layer or regularization purposes. L2 regularization is also implemented in each of the 2D convolutional layers. The final set of layers in the network consist of the flatten and fully connected layers.
Early stopping and reduction of learning rate is applied during the training process. The model performs well with over 90% accuracy, and good precision, recall and f1 scores.